This toolbox provides implementation of multiple Kernel Adaptive Filters in Python
The following are the filters currently available in this toolbox:
- Kernel Least Mean Square (KLMS)
- Quantized Kernel Least Mean Square (Q-KLMS)
- Kernel Maximum Correntropy (KMCC)
- Quantized Kernel Maximum Correntropy (Q-KMCC)
In order for this toolbox to run, the following are the requirements:
- Python 3.4+
- Numpy -
pip3 install numpy - Numexpr -
pip3 install numexpr - SciPy -
pip3 install scipy - Scikit-Learn -
pip3 install sklearn - Matplotlib -
pip3 install matplotlib
To apply the models in code, first import the models.py module.
import models.py
Then initialise any of the desired models by creating an instance of the model class.
klms = models.klms(first_input, first_output, sigma, learning_rate)
qklms = models.qklms(first_input, first_output, threshold_distance, sigma, learning_rate)
kmcc = models.kmcc(first_input, first_output, sigma, learning_rate)
qkmcc = models.qkmcc(first_input, first_output, threshold_distance, sigma, learning_rate)
As these are Online Learning Methods, as data is available, the respective models are updated using the updated function with the new input and output data.
model.update(new_input, new_output)
(Here model is the previously initialised klms, qklms, kmcc, qkmcc.)
The predictions of the model are available with the pred instance variable and the coefficients/weights of the model with the weights instance variable.
predictions = model.pred
weights = model.weights
These values are updated every time the update method is called.